Network Working Group                                       B. Callaghan
Request for Comments: 2055                        Sun Microsystems, Inc.
Category: Informational                                     October 1996


                     WebNFS Server Specification

Status of this Memo

  This memo provides information for the Internet community.  This memo
  does not specify an Internet standard of any kind.  Distribution of
  this memo is unlimited.

Abstract

  This document describes the specifications for a server of WebNFS
  clients.  WebNFS extends the semantics of versions 2 and 3 of the NFS
  protocols to allow clients to obtain filehandles more easily, without
  recourse to the portmap or MOUNT protocols.  In removing the need for
  these protocols, WebNFS clients see benefits in faster response to
  requests, easy transit of firewalls and better server scalability
  This description is provided to facilitate compatible implementations
  of WebNFS servers.

Table of Contents

  1.    Introduction . . . . . . . . . . . . . . . . . . . . . . . 2
  2.    TCP vs UDP . . . . . . . . . . . . .   . . . . . . . . . . 2
  3.    Well-known Port  . . . . . . . . . . . . . . . . . . . . . 2
  4.    Server Port Monitoring . . . . . . . . . . . . . . . . . . 3
  5.    Public Filehandle  . . . . . . . . . . . . . . . . . . . . 3
  5.1     Version 2 Public Filehandle  . . . . . . . . . . . . . . 3
  5.2     Version 3 Public Filehandle  . . . . . . . . . . . . . . 4
  6.    Multi-component Lookup . . . . . . . . . . . . . . . . . . 4
  6.1     Canonical Path vs. Native Path . . . . . . . . . . . . . 5
  6.2     Symbolic Links . . . . . . . . . . . . . . . . . . . . . 6
  6.3     Export Spanning Pathnames  . . . . . . . . . . . . . . . 6
  7.    Location of Public Filehandle  . . . . . . . . . . . . . . 7
  8.    Index Files  . . . . . . . . . . . . . . . . . . . . . . . 7
  9.    Bibliography . . . . . . . . . . . . . . . . . . . . . . . 8
  10.   Security Considerations  . . . . . . . . . . . . . . . . . 9
  11.   Acknowledgements . . . . . . . . . . . . . . . . . . . . . 9
  12.   Author's Address . . . . . . . . . . . . . . . . . . . . . 10








Callaghan                    Informational                      [Page 1]

RFC 2055              WebNFS Server Specification           October 1996


1. Introduction

  The NFS protocol provides access to shared filesystems across
  networks.  It is intended to be machine, operating system, network
  architecture, and transport independent.  The protocol currently
  exists in two versions: version 2 [RFC1094] and version 3 [RFC1813],
  both built on Sun RPC [RFC1831] and its associated eXternal Data
  Representation (XDR) [RFC1832]. This document assumes some
  familiarity with the NFS protocol and underlying RPC protocols.

  WebNFS servers implement semantic extensions to both versions of the
  NFS protocol to support a lightweight binding mechanism for
  conventional or web browser clients that need to communicate with NFS
  servers across the Internet. a WebNFS server supports the public
  filehandle and multi-component lookup features described herein, as
  well as meeting some additional requirements.

  For a description of WebNFS client requirements, read RFC 2054.

2. TCP vs UDP

  The NFS protocol is most well known for its use of UDP which performs
  acceptably on local area networks.  However, on wide area networks
  with error prone, high-latency connections and bandwidth contention,
  TCP is well respected for its congestion control and superior error
  handling.  A growing number of NFS implementations now support the
  NFS protocol over TCP connections.

  A WebNFS client will first attempt to connect to its server with a
  TCP connection.  If the server refuses the connection, the client
  will attempt to use UDP.  All WebNFS servers should support client
  use of TCP and must support UDP.

3. Well-known Port

  While Internet protocols are generally identified by registered port
  number assignments, RPC based protocols register a 32 bit program
  number and a dynamically assigned port with the portmap service which
  is registered on the well-known port 111.  Since the NFS protocol is
  RPC-based, NFS servers register their port assignment with the
  portmap service.

  NFS servers are constrained by a requirement to re-register at the
  same port after a server crash and recovery so that clients can
  recover simply by retransmitting an RPC request until a response is
  received.  This is simpler than the alternative of having the client
  repeatedly check with the portmap service for a new port assignment.
  NFS servers typically achieve this port invariance by registering a



Callaghan                    Informational                      [Page 2]

RFC 2055              WebNFS Server Specification           October 1996


  constant port assignment, 2049, for both UDP and TCP.

  To avoid the overhead of contacting the server's portmap service, and
  to facilitate transit through packet filtering firewalls, WebNFS
  clients optimistically assume that WebNFS servers register on port
  2049.  Most NFS servers use this port assignment already, so this
  client optimism is well justified.

  A WebNFS server must register on UDP port 2049 and TCP port 2049 if
  TCP is supported.

4. Server Port Monitoring

  Some NFS servers accept requests only from reserved UDP or TCP ports,
  i.e. port numbers below 1024.  These "privileged" ports are available
  only to Unix processes with superuser permissions.  Requests that do
  not originate from the range of reserved ports are rejected.  This an
  optimistic way of preventing direct access to the server from user
  processes that may attempt to spoof AUTH_UNIX RPC credentials.

  Since WebNFS clients are not required to use reserved ports, a WebNFS
  server must not check the originating port for requests to
  filesystems made available to WebNFS clients.

5. Public Filehandle

  The public filehandle is an NFS file handle with a reserved value and
  special semantics that allow an initial filehandle to be obtained.  A
  WebNFS client can use the public filehandle as an initial filehandle
  without using the MOUNT protocol.  Since NFS version 2 and version 3
  have different filehandle formats, the public filehandle is defined
  differently for each.

  The public filehandle is a zero filehandle.  For NFS version 2 this
  is a filehandle with 32 zero octets.  A version 3 public filehandle
  has zero length.

5.1 Version 2 Public Filehandle

  A version 2 filehandle is defined in RFC1094 as an opaque value
  occupying 32 octets.  A version 2 public filehandle has a zero in
  each octet, i.e. all zeros.

   1                                                             32
  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  |0|0|0|0|0|0|0|0|0|0|0|0|0|0|0|0|0|0|0|0|0|0|0|0|0|0|0|0|0|0|0|0|
  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+




Callaghan                    Informational                      [Page 3]

RFC 2055              WebNFS Server Specification           October 1996


5.2 Version 3 Public Filehandle

  A version 3 filehandle is defined in RFC1813 as a variable length
  opaque value occupying up to 64 octets.  The length of the filehandle
  is indicated by an integer value contained in a 4 octet value which
  describes the number of valid octets that follow. A version 3 public
  filehandle has a length of zero.

  +-+-+-+-+
  |   0   |
  +-+-+-+-+

6. Multi-component Lookup

  Normally the NFS LOOKUP request (versions 2 or 3) takes a directory
  file handle along with the name of a directory member, and returns
  the filehandle of the directory member.  If a client needs to
  evaluate a pathname that contains a sequence of components, then
  beginning with the directory file handle of the first component it
  must issue a series of LOOKUP requests one component at a time.  For
  instance, evaluation of the Unix path "a/b/c" will generate separate
  LOOKUP requests for each component of the pathname "a", "b", and "c".

  A LOOKUP request that uses the public file handle can provide a
  pathname containing multiple components.  The server is expected to
  evaluate the entire pathname and return a filehandle for the final
  component. The pathname syntax is assumed to be understood by the
  server, but the client must not make assumptions of the pathname
  syntax.

  A Unix server, for instance, uses a slash "/" character to separate
  components in a Unix pathname.

  For example, rather than evaluate the path "a/b/c" as:

       LOOKUP  FH=0x0  "a"  --->
                            <---  FH=0x1
       LOOKUP  FH=0x1  "b"  --->
                            <---  FH=0x2
       LOOKUP  FH=0x2  "c"  --->
                            <---  FH=0x3

  Relative to the public filehandle these three LOOKUP requests can be
  replaced by a single multi-component lookup:

       LOOKUP  FH=0x0  "a/b/c"  --->
                                <---  FH=0x3




Callaghan                    Informational                      [Page 4]

RFC 2055              WebNFS Server Specification           October 1996


  Multi-component lookup is supported only for LOOKUP requests relative
  to the public filehandle.

6.1 Canonical Path vs. Native Path

  If the pathname in a multi-component LOOKUP request begins with a
  printable ASCII character, then it must be a canonical path.  A
  canonical path is a hierarchically-related, slash-separated sequence
  of components, <directory>/<directory>/.../<name>.

  Occurrences of the "/" character within a component will be escaped
  using the escape code %2f.  Non-printable ascii characters (with
  values in the range 00-1F and 7f hexadecimal) will also be escaped
  using the "%" character to introduce a two digit hexadecimal code.
  Occurrences of the "%" character that do not introduce an encoded
  character will themselves be encoded with %25.

  If the first character of a canonical path is a slash, then the
  canonical path must be evaluated relative to the server's root
  directory.  If the first character is not a slash, then the path must
  be evaluated relative to the directory with which the public
  filehandle is associated.

  Not all WebNFS servers can support arbitrary use of absolute paths.
  Clearly, the server cannot return a filehandle if the path identifies
  a file or directory that is not exported by the server.  In addition,
  some servers will not return a filehandle if the path names a file or
  directory in an exported filesystem different from the one that is
  associated with the public filehandle.

  If the first character of the path is 0x80 (non-ascii) then the
  following character is the first in a native path.  A native path
  conforms to the natural pathname syntax of the server. For example:

       Lookup for Canonical Path:

               LOOKUP FH=0x0 "/a/b/c"

       Lookup for Native Path:

               LOOKUP FH=0x0  0x80 "a:b:c"

  Other introductory characters in the range 0x81 - 0xff may be added
  in future specifications.  If the server receives any character in
  this range that it does not understand then it must return an error
  to the client: NFSERR_IO for NFS V2, NFS3ERR_IO for NFS V3.





Callaghan                    Informational                      [Page 5]

RFC 2055              WebNFS Server Specification           October 1996


6.2 Symbolic Links

  Servers that support symbolic links may encounter pathname components
  that are symbolic links.  The server is expected to evaluate these
  symbolic links as a part of the normal pathname evaluation process.
  This is a different semantic from that of conventional component-at-
  a-time pathname evaluation by NFS clients, where the client is
  expected to do the evaluation.

  However, if the final component is a symbolic link, the server must
  return its filehandle and let the client evaluate it.

6.3 Export Spanning Pathnames

  The server may evaluate a pathname, either through a multi-component
  LOOKUP or as a symbolic link embedded in a pathname, that references
  a file or directory outside of the exported hierarchy.

  Clearly, if the destination of the path is not in an exported
  filesystem, then the server must return an error to the client.

  Many NFS server implementations rely on the MOUNT protocol for
  checking access to exported filesystems and NFS server does no access
  checking.  The NFS server assumes that the filehandle does double
  duty: identifying a file as well as being a security token. Since
  WebNFS clients do not normally use the MOUNT protocol, a server that
  relies on MOUNT checking cannot automatically grant access to another
  exported filesystem at the destination of a spanning path. These
  servers must return an error.

  For example: the server exports two filesystems.  One is associated
  with the public filehandle.

     /export/this   (public filehandle)

     /export/that

  The server receives a LOOKUP request with the public filehandle that
  identifies a file or directory in the other exported filesystem:

     LOOKUP 0x0  "../that/file"
  or
     LOOKUP 0x0  "/export/that/file"

  Even though the pathname destination is in an exported filesystem,
  the server cannot return a filehandle without an assurance that the
  client's use of this filehandle will be authorized.




Callaghan                    Informational                      [Page 6]

RFC 2055              WebNFS Server Specification           October 1996


  Servers that check client access to an export on every NFS request
  have more flexibility.  These servers can return filehandles for
  paths that span exports since the client's use of the filehandle for
  the destination filesystem will be checked by the NFS server.

7. Location of Public Filehandle

  A server administrator can associate the public filehandle with any
  file or directory. For instance, a WebNFS server administrator could
  attach the public filehandle to the root of an anonymous FTP archive,
  so that anonymous FTP pathnames could be used to identify files in
  the FTP hierarchy, e.g.

     # share -o ro,public  /export/ftp

  On servers that support spanning paths, the public filehandle need
  not necessarily be attached to an exported directory, though a
  successful LOOKUP relative to the public filehandle must identify a
  file or directory that is exported.

  For instance, if an NFS server exports a directory "/export/foo" and
  the public filehandle is attached to the server's root directory,
  then a LOOKUP of "export/foo" relative to the public filehandle will
  return a valid file handle but a LOOKUP of "export" will return an
  access error since the server's "/export" directory is not exported.

              /            (public filehandle is here)
             /\
            /  \
           /   export      (not exported)
          /    /\
         /    /  \
        /    /   foo       (exported)


     LOOKUP 0x0  "export"      (returns an error)

     LOOKUP 0x0  "export/foo"  (returns an filehandle)

8. Index Files

  Most HTTP servers support the concept of an index file.  If a browser
  references a directory that contains an index file, then the server
  will return the contents of the index file rather than a directory
  listing.  For instance if a browser requests "a/b/c" then the server
  might return the contents of "a/b/c/index.html".





Callaghan                    Informational                      [Page 7]

RFC 2055              WebNFS Server Specification           October 1996


  A WebNFS server may choose to emulate this feature for the benefit of
  clients using the NFS protocol to browse a Web hierarchy. On
  receiving a multi-component lookup for a canonical path that names a
  directory, the server can check that directory for the presence of an
  index file.  If the file exists then the server may choose to return
  the filehandle of the index file instead of the directory.  Index
  files are commonly called "index.html" though the name is usually
  configurable.

9. Bibliography

  [RFC1831]       Srinivasan, R., "RPC: Remote Procedure Call
                  Protocol Specification Version 2", RFC 1831,
                  August 1995.
                  http://www.internic.net/rfc/rfc1831.txt

  [RFC1832]       Srinivasan, R., "XDR: External Data Representation
                  Standard," RFC 1832, August 1995.
                  http://www.internic.net/rfc/rfc1832.txt

  [RFC1833]       Srinivasan, R., "Binding Protocols for ONC RPC
                  Version 2", RFC 1833, August 1995.
                  http://www.internic.net/rfc/rfc1833.txt

  [RFC1094]       Sun Microsystems, Inc., "Network Filesystem
                  Specification", RFC 1094, March 1989.  NFS
                  version 2 protocol specification.
                  http://www.internic.net/rfc/rfc1094.txt

  [RFC1813]       Sun Microsystems, Inc., "NFS Version 3 Protocol
                  Specification", RFC 1813, June 1995.  NFS version
                  3 protocol specification.
                  http://www.internic.net/rfc/rfc1813.txt

  [RFC2054]       Callaghan, B., "WebNFS Client Specification",
                  RFC 2054, October 1996.
                  http://www.internic.net/rfc/rfc2054.txt

  [Sandberg]      Sandberg, R., D. Goldberg, S. Kleiman, D. Walsh,
                  B.  Lyon, "Design and Implementation of the Sun
                  Network Filesystem," USENIX Conference
                  Proceedings, USENIX Association, Berkeley, CA,
                  Summer 1985.  The basic paper describing the
                  SunOS implementation of the NFS version 2
                  protocol, and discusses the goals, protocol
                  specification and trade-offs.





Callaghan                    Informational                      [Page 8]

RFC 2055              WebNFS Server Specification           October 1996


  [X/OpenNFS]     X/Open Company, Ltd., X/Open CAE Specification:
                  Protocols for X/Open Internetworking: XNFS,
                  X/Open Company, Ltd., Apex Plaza, Forbury Road,
                  Reading Berkshire, RG1 1AX, United Kingdom,
                  1991.  This is an indispensable reference for
                  NFS version 2 protocol and accompanying
                  protocols, including the Lock Manager and the
                  Portmapper.

  [X/OpenPCNFS]   X/Open Company, Ltd., X/Open CAE Specification:
                  Protocols for X/Open Internetworking: (PC)NFS,
                  Developer's Specification, X/Open Company, Ltd.,
                  Apex Plaza, Forbury Road, Reading Berkshire, RG1
                  1AX, United Kingdom, 1991.  This is an
                  indispensable reference for NFS version 2
                  protocol and accompanying protocols, including
                  the Lock Manager and the Portmapper.

10. Security Considerations

  Since the WebNFS server features are based on NFS protocol versions 2
  and 3, the RPC security considerations described in RFC 1094, RFC
  1813, and Appendix A of RFC 1831 apply here also.

  Clients and servers may separately negotiate secure connection
  schemes for authentication, data integrity, and privacy.

  Implementors must note carefully the implications of export spanning
  pathnames as described in section 6.3.

11. Acknowledgements

  This specification was extensively reviewed by the NFS group at
  SunSoft and brainstormed by Michael Eisler.

















Callaghan                    Informational                      [Page 9]

RFC 2055              WebNFS Server Specification           October 1996


12. Author's Address

  Address comments related to this document to:

  [email protected]


  Brent Callaghan
  Sun Microsystems, Inc.
  2550 Garcia Avenue
  Mailstop Mpk17-201
  Mountain View, CA 94043-1100

  Phone: 1-415-786-5067
  Fax:   1-415-786-5896
  EMail: [email protected]



































Callaghan                    Informational                     [Page 10]